@elizaos/plugin-ollama 2.0.0-beta.1 → 2.0.3-beta.3

package/dist/node/index.node.js

@@ -9,7 +9,7 @@ import { createOllama } from "ollama-ai-provider-v2";
 // utils/config.ts
 var DEFAULT_OLLAMA_URL = "http://localhost:11434";
 var DEFAULT_SMALL_MODEL = "eliza-1-2b";
-var DEFAULT_LARGE_MODEL = "eliza-1-9b";
+var DEFAULT_LARGE_MODEL = "eliza-1-4b";
 var DEFAULT_EMBEDDING_MODEL = "eliza-1-2b";
 function getEnvValue(key) {
   if (typeof process === "undefined" || !process.env) {
@@ -21,12 +21,12 @@ function getEnvValue(key) {
 function getSetting(runtime, key, defaultValue) {
   const value = runtime.getSetting(key);
   if (value !== undefined && value !== null) {
-    return String(value);
+    return String(value).trim();
   }
-  return getEnvValue(key) ?? defaultValue;
+  return getEnvValue(key)?.trim() ?? defaultValue;
 }
 function getBaseURL(runtime) {
-  const apiEndpoint = getSetting(runtime, "OLLAMA_API_ENDPOINT") || getSetting(runtime, "OLLAMA_API_URL") || DEFAULT_OLLAMA_URL;
+  const apiEndpoint = getSetting(runtime, "OLLAMA_API_ENDPOINT") || getSetting(runtime, "OLLAMA_API_URL") || getSetting(runtime, "OLLAMA_BASE_URL") || DEFAULT_OLLAMA_URL;
   if (!apiEndpoint.endsWith("/api")) {
     return apiEndpoint.endsWith("/") ? `${apiEndpoint}api` : `${apiEndpoint}/api`;
   }
@@ -174,39 +174,53 @@ async function ensureModelAvailable(model, providedBaseURL, customFetch) {
 }
 
 // models/embedding.ts
+var INIT_PROBE_TEXT = "dimension probe";
+function extractText(params) {
+  if (params === null) {
+    return null;
+  }
+  if (typeof params === "string") {
+    return params;
+  }
+  if (typeof params === "object" && typeof params.text === "string") {
+    return params.text;
+  }
+  throw new Error("Invalid input format for embedding: expected string or { text: string }");
+}
 async function handleTextEmbedding(runtime, params) {
+  const text = extractText(params);
+  const isInitProbe = text === null;
+  if (!isInitProbe && !text.trim()) {
+    throw new Error("Cannot generate embedding for empty text");
+  }
   try {
     const baseURL = getBaseURL(runtime);
     const customFetch = runtime.fetch ?? undefined;
     const ollama = createOllama({
-      fetch: customFetch,
+      ...customFetch ? { fetch: customFetch } : {},
       baseURL
     });
     const modelName = getEmbeddingModel(runtime);
     logger2.log(`[Ollama] Using TEXT_EMBEDDING model: ${modelName}`);
     await ensureModelAvailable(modelName, baseURL, customFetch);
-    let text = typeof params === "string" ? params : params ? params.text || "" : "";
     const maxChars = 8000 * 4;
-    if (text.length > maxChars) {
-      logger2.warn(`[Ollama] Embedding input too long (~${Math.ceil(text.length / 4)} tokens), truncating to ~8000 tokens`);
-      text = text.slice(0, maxChars);
+    let embeddingText = isInitProbe ? INIT_PROBE_TEXT : text;
+    if (embeddingText.length > maxChars) {
+      logger2.warn(`[Ollama] Embedding input too long (~${Math.ceil(embeddingText.length / 4)} tokens), truncating to ~8000 tokens`);
+      embeddingText = embeddingText.slice(0, maxChars);
     }
-    const embeddingText = text || "test";
-    try {
-      const embedParams = {
-        model: ollama.embedding(modelName),
-        value: embeddingText
-      };
-      const { embedding, usage } = await embed(embedParams);
+    const embedParams = {
+      model: ollama.embedding(modelName),
+      value: embeddingText
+    };
+    const { embedding, usage } = await embed(embedParams);
+    if (!isInitProbe) {
       emitModelUsed(runtime, ModelType.TEXT_EMBEDDING, modelName, normalizeTokenUsage(usage) ?? estimateEmbeddingUsage(embeddingText));
-      return embedding;
-    } catch (embeddingError) {
-      logger2.error({ error: embeddingError }, "Error generating embedding");
-      return Array(1536).fill(0);
     }
+    return embedding;
   } catch (error) {
     logger2.error({ error }, "Error in TEXT_EMBEDDING model");
-    return Array(1536).fill(0);
+    throw error instanceof Error ? error : new Error(String(error));
   }
 }
 
@@ -500,11 +514,11 @@ function firstString(...values) {
 }
 
 // models/text.ts
-var TEXT_NANO_MODEL_TYPE = ModelType2.TEXT_NANO ?? "TEXT_NANO";
-var TEXT_MEDIUM_MODEL_TYPE = ModelType2.TEXT_MEDIUM ?? "TEXT_MEDIUM";
-var TEXT_MEGA_MODEL_TYPE = ModelType2.TEXT_MEGA ?? "TEXT_MEGA";
-var RESPONSE_HANDLER_MODEL_TYPE = ModelType2.RESPONSE_HANDLER ?? "RESPONSE_HANDLER";
-var ACTION_PLANNER_MODEL_TYPE = ModelType2.ACTION_PLANNER ?? "ACTION_PLANNER";
+var TEXT_NANO_MODEL_TYPE = ModelType2.TEXT_NANO;
+var TEXT_MEDIUM_MODEL_TYPE = ModelType2.TEXT_MEDIUM;
+var TEXT_MEGA_MODEL_TYPE = ModelType2.TEXT_MEGA;
+var RESPONSE_HANDLER_MODEL_TYPE = ModelType2.RESPONSE_HANDLER;
+var ACTION_PLANNER_MODEL_TYPE = ModelType2.ACTION_PLANNER;
 function summarizeAiSdkErrorForLogs(error, depth = 0) {
   if (depth > 4) {
     return { note: "max depth summarizing nested error" };
@@ -565,7 +579,7 @@ function serializeStructuredGenerateTextResult(result) {
   if (result.output !== undefined && result.output !== null) {
     return typeof result.output === "string" ? result.output : JSON.stringify(result.output);
   }
-  const trimmed = result.text?.trim() ?? "";
+  const trimmed = result.text.trim();
   if (trimmed)
     return trimmed;
   throw new Error("[Ollama] Structured generation returned no text or output.");
@@ -574,7 +588,7 @@ function buildNativeResultCast(result, modelName, usage) {
   const payload = {
     text: result.text,
     toolCalls: mapAiSdkToolCallsToCore(result.toolCalls),
-    finishReason: String(result.finishReason ?? ""),
+    finishReason: String(result.finishReason),
     usage,
     providerMetadata: { modelName }
   };
@@ -588,9 +602,7 @@ function buildOllamaStreamTextResult(args) {
   });
   const usagePromise = Promise.resolve(streamResult.usage).then(async (usage) => {
     const fullText = await textPromise;
-    const normalized = normalizeTokenUsage(usage) ?? estimateUsage(args.promptForEstimate, fullText);
-    emitModelUsed(args.runtime, args.modelType, args.model, normalized);
-    return normalized;
+    return normalizeTokenUsage(usage) ?? estimateUsage(args.promptForEstimate, fullText);
   }).catch(() => {
     return;
   });
@@ -606,9 +618,12 @@ function buildOllamaStreamTextResult(args) {
       throw streamErr;
     } finally {
       if (completed) {
-        await usagePromise.catch(() => {
+        const usage = await usagePromise.catch(() => {
           return;
         });
+        if (usage) {
+          emitModelUsed(args.runtime, args.modelType, args.model, usage);
+        }
       }
     }
   }
@@ -634,9 +649,7 @@ function buildOllamaStreamWithToolsResult(args) {
   const toolCallsPromise = Promise.resolve(streamResult.toolCalls).then((calls) => mapAiSdkToolCallsToCore(calls)).catch(() => []);
   const usagePromise = Promise.resolve(streamResult.usage).then(async (usage) => {
     const fullText = await sdkTextPromise;
-    const normalized = normalizeTokenUsage(usage) ?? estimateUsage(args.promptForEstimate, fullText);
-    emitModelUsed(args.runtime, args.modelType, args.model, normalized);
-    return normalized;
+    return normalizeTokenUsage(usage) ?? estimateUsage(args.promptForEstimate, fullText);
   }).catch(() => {
     return;
   });
@@ -657,6 +670,11 @@ function buildOllamaStreamWithToolsResult(args) {
         const first = mapped[0];
         if (first) {
           yield stringifyPlannerToolArgs(first.arguments);
+        } else {
+          const fallbackText = await sdkTextPromise;
+          if (fallbackText) {
+            yield fallbackText;
+          }
         }
       } else {
         for await (const chunk of streamResult.textStream) {
@@ -669,9 +687,12 @@ function buildOllamaStreamWithToolsResult(args) {
       throw streamErr;
     } finally {
       if (completed) {
-        await usagePromise.catch(() => {
+        const usage = await usagePromise.catch(() => {
           return;
         });
+        if (usage) {
+          emitModelUsed(args.runtime, args.modelType, args.model, usage);
+        }
       }
     }
   }
@@ -705,26 +726,21 @@ function getModelNameForType(runtime, modelType) {
 }
 async function handleTextWithModelType(runtime, modelType, params) {
   const extended = params;
-  const structuredDisabled = isOllamaStructuredOutputDisabled(runtime);
-  let responseSchema = extended.responseSchema;
-  if (structuredDisabled && extended.responseSchema) {
-    logger3.debug("[Ollama] OLLAMA_DISABLE_STRUCTURED_OUTPUT is set — ignoring responseSchema for this call.");
-    responseSchema = undefined;
-  }
-  const tools = normalizeNativeTools(extended.tools);
-  const {
-    prompt,
-    maxTokens = 8192,
-    temperature = 0.7,
-    frequencyPenalty = 0.7,
-    presencePenalty = 0.7
-  } = params;
+  const { prompt, temperature = 0.7, frequencyPenalty = 0.7, presencePenalty = 0.7 } = params;
+  const maxTokens = params.omitMaxTokens ? undefined : params.maxTokens ?? 8192;
   let modelIdForLog = "";
   try {
+    const structuredDisabled = isOllamaStructuredOutputDisabled(runtime);
+    let responseSchema = extended.responseSchema;
+    if (structuredDisabled && extended.responseSchema) {
+      logger3.debug("[Ollama] OLLAMA_DISABLE_STRUCTURED_OUTPUT is set — ignoring responseSchema for this call.");
+      responseSchema = undefined;
+    }
+    const tools = normalizeNativeTools(extended.tools);
     const baseURL = getBaseURL(runtime);
     const customFetch = runtime.fetch ?? undefined;
     const ollama = createOllama2({
-      fetch: customFetch,
+      ...customFetch ? { fetch: customFetch } : {},
       baseURL
     });
     const model = getModelNameForType(runtime, modelType);
@@ -746,7 +762,7 @@ async function handleTextWithModelType(runtime, modelType, params) {
     const toolChoice = tools ? normalizeToolChoice(extended.toolChoice) : undefined;
     const shouldReturnNative = Boolean(hasChatMessages || tools || extended.toolChoice || outputSpec !== undefined);
     const renderedPrompt = hasChatMessages ? "" : renderChatMessagesForPrompt(params.messages, {
-      omitDuplicateSystem: system
+      ...system ? { omitDuplicateSystem: system } : {}
     }) ?? prompt ?? "";
     const promptOrMessages = hasChatMessages ? { messages: normalizedMessages } : { prompt: renderedPrompt };
     const resolvedStopSequences = Array.isArray(params.stopSequences) && params.stopSequences.length > 0 ? params.stopSequences : undefined;
@@ -756,9 +772,9 @@ async function handleTextWithModelType(runtime, modelType, params) {
       ...promptOrMessages,
       system,
       temperature,
-      maxOutputTokens: maxTokens,
       frequencyPenalty,
       presencePenalty,
+      ...typeof maxTokens === "number" ? { maxOutputTokens: maxTokens } : {},
       ...resolvedStopSequences ? { stopSequences: resolvedStopSequences } : {},
       ...tools ? { tools, ...toolChoice ? { toolChoice } : {} } : {},
       ...outputSpec ? { output: outputSpec } : {}
@@ -806,7 +822,7 @@ async function handleTextWithModelType(runtime, modelType, params) {
       endpoint = getBaseURL(runtime);
     } catch {}
     logOllamaTextFailure("generateText", String(modelType), modelIdForLog || "(unknown)", endpoint, error);
-    return "Error generating text. Please try again later.";
+    throw error;
   }
 }
 async function handleTextSmall(runtime, params) {
@@ -841,11 +857,11 @@ function getProcessEnv() {
   return process.env;
 }
 var env = getProcessEnv();
-var TEXT_NANO_MODEL_TYPE2 = ModelType3.TEXT_NANO ?? "TEXT_NANO";
-var TEXT_MEDIUM_MODEL_TYPE2 = ModelType3.TEXT_MEDIUM ?? "TEXT_MEDIUM";
-var TEXT_MEGA_MODEL_TYPE2 = ModelType3.TEXT_MEGA ?? "TEXT_MEGA";
-var RESPONSE_HANDLER_MODEL_TYPE2 = ModelType3.RESPONSE_HANDLER ?? "RESPONSE_HANDLER";
-var ACTION_PLANNER_MODEL_TYPE2 = ModelType3.ACTION_PLANNER ?? "ACTION_PLANNER";
+var TEXT_NANO_MODEL_TYPE2 = ModelType3.TEXT_NANO;
+var TEXT_MEDIUM_MODEL_TYPE2 = ModelType3.TEXT_MEDIUM;
+var TEXT_MEGA_MODEL_TYPE2 = ModelType3.TEXT_MEGA;
+var RESPONSE_HANDLER_MODEL_TYPE2 = ModelType3.RESPONSE_HANDLER;
+var ACTION_PLANNER_MODEL_TYPE2 = ModelType3.ACTION_PLANNER;
 var ollamaPlugin = {
   name: "ollama",
   description: "Ollama plugin for local LLM inference",
@@ -885,7 +901,8 @@ var ollamaPlugin = {
       }
     }
     try {
-      const response = await fetch(`${apiBase}/api/tags`, {
+      const fetchImpl = runtime.fetch ?? fetch;
+      const response = await fetchImpl(`${apiBase}/api/tags`, {
         method: "GET",
         headers: { "Content-Type": "application/json" }
       });
@@ -1059,5 +1076,5 @@ export {
   DEFAULT_EMBEDDING_MODEL
 };
 
-//# debugId=28C4D03399E00E1164756E2164756E21
+//# debugId=ADFD08073B38308364756E2164756E21
 //# sourceMappingURL=index.node.js.map